This article exploits the complex sequential structure of the diary data in the American Heritage Time Use Study (AHTUS) and constructs three classes of indicators that capture the quality of leisure (pure leisure, co-present leisure, and leisure fragmentation) to show that the relative growth in leisure time enjoyed by low-educated individuals documented in previous studies has been accompanied by a relative decrease in the quality of that leisure time. These results are not driven by any single leisure activity, such as time spent watching television. Our findings may offer a more comprehensive picture of inequality in the United States and provide a basis for weighing the relative decline in earnings and consumption for the less-educated against the simultaneous relative growth of leisure.
The basis on which good repute in any highly organized industrial community ultimately rests is pecuniary strength; and the means of showing pecuniary strength, and so of gaining or retaining a good name, are leisure and a conspicuous consumption of goods.Veblen (1953)
The distribution of leisure in the United States over the past four decades contrasts remarkably with the evolution of inequality in wages and expenditure over the same period of time. Despite growing wage and expenditure inequality in the United States (e.g., Attanasio and Davis 1996; Katz and Autor 1999; Krueger and Perri 2006), the cross-sectional distribution of leisure time expanded over the past 40 years (Aguiar and Hurst 2007). Although the level of leisure in 1965 was roughly equal across educational groups, the subsequent increase in leisure was largest for low-educated adults. Highly educated individuals now have substantially less leisure time than low-educated individuals.1 This variation in leisure across educational levels has also been documented by Gershuny (2009a), who found a reversal of the previously negative relationship between human capital and work time in six developed economies.
In this article, we look inside the black box of leisure time, exploiting the rich information in the American Heritage Time Use Study (AHTUS) time-use diary data to construct three classes of indicators that capture the quality of leisure. Most studies using the AHTUS focus their analysis merely on the primary activity field of the diary instrument. We deploy more of the rich diary information in the U.S. time-use series—secondary activities and co-presence—to measure the distribution of quantity and quality of leisure. These diary records allow us to look at the simultaneity of leisure activities with nonleisure activities, the presence of other individuals while the respondent is engaging in a leisure activity, and the extent to which leisure events are interrupted by other activities (i.e., pure leisure, co-present leisure, and leisure fragmentation).
The majority of our leisure-quality indicators show that despite increases in the quantity of leisure over this period (as reported by, e.g., Aguiar and Hurst (2007)), the quality of leisure decreased for all groups. This decline in leisure quality is consistent with the results using instant enjoyment data showing declines in the time spent in the sorts of activities labeled “enjoyable and engaging forms of leisure” (Krueger 2007). Despite general increases in leisure time, Americans report feeling increasingly harried now compared with 40 years ago (e.g., Hamermesh and Lee 2007; Robinson and Godbey 1997; Schor 1993). Our findings may help explain this paradox.
More directly relevant to the focus of this article is a comparison across educational groups over time. It emerges that qualitative differences in leisure time across educational groups partially compensate for highly educated individuals now having less leisure time (e.g., Aguiar and Hurst 2007; Gimenez-Nadal and Sevilla-Sanz 2012). Leisure increased nearly an hour per week more for low-educated (with at most a high school diploma) men than for highly educated (with some college or more) men, and 3.5 hour more for low-educated women than for highly educated women. However, pure leisure (the amount of leisure time that is not infringed upon by nonleisure activities) declined substantially more for low-educated individuals than for highly educated individuals over the period for which we can observe it. Between 1965 and 1985, low-educated men experienced a decline of 1 hour and 35 minutes per week versus 50 minutes for highly educated men. Similarly, low-educated women experienced a decline in pure leisure of 1 hour and 50 minutes versus a decline of just 1 hour for highly educated women over the same period. Between 1965 and 2003, leisure time spent in the company of the spouse declined 1 hour for men with at most a high school diploma, whereas men with some college or more did not experience a statistically significant decrease in this indicator. Similarly, also between 1965 and 2003, the number of hours low-educated men and women spent in the company of other adults decreased to a greater extent than that of highly educated adults. Highly educated men also experienced a more favorable trend in leisure fragmentation than low-educated men. In contrast, highly educated women—and, in particular, highly educated working women—have experienced a less-favorable trend in leisure fragmentation than their low-educated counterparts.
We find that no single leisure activity can explain the differential trends in terms of leisure quantity and quality across educational groups over this period. We show that the pattern of change of leisure quality (and quantity) across educational groups remains very similar when television watching is not included in the definition of leisure. Other leisure activities, such as at-home leisure and read/listen, which also represent a significant fraction of the total amount of leisure time, similarly fail to explain the unequal distribution of leisure quality across educational groups.
Our work expands the existing literature on measuring changes in the allocation of time in the United States. The literature mainly concentrates on the study of aggregate totals of time (e.g., Aguiar and Hurst 2007; Ghez and Becker 1975; Juster and Stafford 1985; Robinson and Godbey 1997). However, introducing other dimensions of time in the analysis of inequality is crucially important. Although the scarcity of leisure time may seem analogous to income poverty—in that both reflect the scarcity of resources—the two concepts, in fact, have different historical dynamics. In a growing economy, the goods constraint relaxes over time, whereas the 24-hours-per-day time constraint does not. However, the time-budget constraint might instead be ameliorated by adjusting the quality of leisure over time.
This article also contributes to a recent broadening of focus from production to the measurement of well-being. For example, Stiglitz et al. (2009) recently proposed a broad range of measures of household economic activity to evaluate quality of life, such as time spent in leisure and the instant enjoyment of leisure activities. Our objective indicators of leisure quality may provide an additional basis for interpreting well-being inequality in the United States and for weighing the relative growth of leisure for the low-educated against the simultaneous decline in relative wages and consumption.
This article is organized as follows. The next section describes the time-use data sets used in the analysis and the conceptualization of the quantity of leisure, and also presents the theoretical and empirical underpinnings for our leisure quality indicators. We then show the main results and look more deeply into how the nature of specific leisure activities may have contributed to the differential trends in the quantity and quality of leisure. The final section offers conclusions.
The American Heritage Time Use Study (1965–2003)
We use the American Heritage Time Use Study (AHTUS) (e.g., Fisher et al. 2011) in our main analysis. The AHTUS is a harmonized data set that covers five decades, from 1965 to 2003, over five time-use surveys. Appendix 1 describes the main sample in our analysis, and Table 4 in Appendix 1 shows the five surveys in the AHTUS as well as the harmonization exercise. The main instrument of all the surveys is an activity diary in which respondents record their activities for a consecutive period of 24 hours. For each respondent there is a diary file made up of a sequence of episodes over the 24-hour span, which allows the inclusion of harmonized information on secondary activity and on who else is present at the time of the activity. The AHTUS also allows us to analyze episode files rather than aggregated files.
The Quantity of Leisure
The conceptualization of leisure, and of time-use categories in general, is usually driven by a systematic, principle-driven approach of distinguishing means from ends. The so-called “third-person criterion” excludes activities that might be carried out by some third party without losing the intended utility for the final consumer, such as watching television. Unfortunately, the third-person criterion involves questionable assumptions such that the enjoyment derived from work can legitimately be ignored, and that all leisure is enjoyable.2 Certain activities, such as sleeping, eating, personal and medical care, or resting, do not fall comfortably into the means-versus-ends classification. These activities cannot be purchased in the market, but they also might not be considered leisure in the sense that they are necessary for life. Nonetheless, some variation in the time spent in these activities may result from conscious choice. For example, Biddle and Hamermesh (1990) showed that sleep time responds to economic incentives, such as wages. Indeed, Gershuny (2009b) used diary reports of enjoyment and showed a decreasing marginal utility of sleep (and of other consumption activities).
Rather than trying to resolve this debate on theoretical grounds, we adopt an empirical approach, exploring four commonly used and nested definitions of leisure, ranging from the narrow (activities designed to yield direct utility, such as entertainment, socializing, active recreation, and general relaxation) to the broad (time spent neither in market production nor in nonmarket production). The various measures tell a consistent story, so for brevity, we present here only the results regarding our narrowest measure of leisure—that is, hours per week devoted to all activities that we cannot pay somebody else to do for us and that are not biological needs (e.g., Burda et al. 2008; Hawrylyshyn 1976, 1977; Walker and Gauger 1973).3
Among the activities included in the leisure category are watching television, sport activities, general out-of-home leisure, and socializing. We exclude volunteer activities from our main definition of leisure because they are classified as work under the third-person criterion (see Hawrylyshyn 1976). Although the classification of time-use activities changes over time, and some activities disappear and new activities emerge (just as in the case of expenditure diary categories), the AHTUS nevertheless provides comparable leisure activities for the different years so that it is still possible to run meaningful comparisons over time of broad time-use categories such as leisure.4
The Quality of Leisure
There are different ways of assessing the quality of leisure. One methodology is to use self-reported measures of how enjoyable activities are, in the spirit of the literature on process benefits and experienced utility. Juster and Stafford (1985) defined process benefits as the “direct subjective consequences from engaging in some activities to the exclusion of others.” Going back to the earliest conceptions of utility, from Jeremy Bentham through Francis Ysidro Edgeworth and Alfred Marshall, the concept of experienced utility has been proposed more recently by Kahneman et al. (2004) to refer to a “continuous hedonic flow of pleasure or pain.”
Both lines of research use time-use diaries together with information on enjoyment to assess individuals’ subjective well-being. The process-benefits approach uses activity enjoyment ratings in which respondents rate on a scale from 0 to 10 how much they generally enjoyed a type of activity (e.g., Juster and Stafford 1985). The information gathered this way offers a global and retrospective interpretation of feelings about activities, although they may not serve as a good predictor of the instantaneous satisfaction experienced in any given instance of the activity (Gershuny and Halpin 1996). The literature on experienced utility has proposed the experience sampling method as a superior way for collecting objective instantaneous enjoyment data. As opposed to the activity enjoyment ratings, the experience sampling method collects information on hedonic experiences (or instant enjoyment) in real time. It has, however, never been applied to a representative population sample because it is extremely burdensome for the respondent.5 Alternative methods of collecting data on hedonic experiences, such as the conventional “yesterday diary” used in time-budget surveys (Szalai 1972) or the day reconstruction method (Kahneman et al. 2004) are less costly to implement. Both methods collect information on how the respondent experienced all or some of his or her activities of the previous day, as described by a time-use diary.6
Whereas historical information on time-use diary records is available from 1965 onward for the United States, only one survey contains information about instant enjoyment for a nationally representative sample. We thus adopt a complementary approach to the aforementioned literature and exploit the rich information in the diary to construct three classes of leisure quality indicators that emerge independently from different strands in the socioeconomic and psychological literature.7 The relationship between quality of leisure and some of these indicators—in particular, those related to the presence of other individuals while the respondent engages in leisure activities—has already been directly established using instant-enjoyment data of the sort proposed by the process-benefits and experienced-utility literature. Using the AHTUS and a closely analogous UK Unilever data set, we present a simple validation exercise of all our indicators by analysis of the available direct evidence of the enjoyability of activities in Appendix 2. The validation exercise suggests that these indicators are capable of conveying important information about leisure quality that cannot be explained by the type of leisure activity alone. Even though we lack additional direct information about how much respondents enjoy engaging in a given activity for the decades being analyzed, our indicators seem to be good instruments for assessing trends in the quality of U.S. leisure time.
The first class of indicator is related to activity density. Respondents frequently engage in more than one activity at the same time. The secondary activity is simultaneous with another identified by the diary respondent as the “main activity,” which may in some way complement or qualify it. The underlying rationale behind this indicator is that leisure activities with no “distracting” accompanying activities will be associated with a higher utility than leisure activities accompanied by a secondary activity (see Bittman and Wajcman 2000; Mattingly and Bianchi 2003). For example, Gimenez-Nadal and Ortega-Lapiedra (2010) showed that the leisure of self-employed men is more often intertwined with market work activities, leading self-employed men to report higher levels of time stress. We define pure leisure as leisure that is reported as a primary activity whose secondary activity is not market work, home production, or personal care, and analyze the proportion of pure leisure out of total leisure. Pure leisure cannot be analyzed for 1993 and 2003 because these surveys collected no separately identified secondary activities.
Co-present Leisure: Leisure With Spouse, and Leisure With Adults
The second class of indicators relates to with whom the leisure activity is performed. Consider, first, leisure with the spouse (or partner). The concept of leisure with spouse draws from the empirical evidence found in the socioeconomic literature on spouses’ synchronization of work and leisure activities. Sullivan (1996a) used a 1985 UK time-use survey, a diary survey including instantaneous enjoyment diary information, to show that partners report higher levels of satisfaction when they synchronize their working schedules (and thus maximize the potential time they can spend in leisure activities together).8 Hamermesh (1999), Hallberg (2003), and Jenkins and Osberg (2005) followed Sullivan (1996a) in finding that synchronization of leisure activities between partners is indeed greater than random male-female pairing would predict. We thus use information on whether leisure as primary activity is carried out while the spouse/partner is present in order to calculate what percentage of total leisure time is leisure with spouse (or partner), as an indicator of leisure quality. Because of demographic changes regarding the propensity to marry, which declined in the United States during this time period because of the delay in the age of marriage and the increase in divorce rates (e.g., Goldstein 1999; Rindfuss et al. 1996), we restrict the sample to those individuals with a partner when computing this indicator. The indicator of leisure with spouse can be constructed for all the surveys except for those in 1985 and 1993, in which information on spouse or partner co-presence was not gathered.
The second indicator in this class uses information on whether another adult was present during a leisure activity to construct the percentage of total leisure that constitutes leisure with adults (i.e., leisure time spent neither alone nor in the presence of children).9 We can calculate this for the 1965, 1975, and 2003 surveys. There is evidence from instant enjoyment data suggesting that individuals report higher levels of instant satisfaction from activities done in the company of others than those done alone (e.g., Helliwell and Putnam 2005; Kahneman et al. 2004). In fact, the adverse effects of isolation on mental health are well known in the epidemiological and psychological literature (e.g., Berkman et al. 2004; Berkman and Glass 2000; Eng et al. 2002; House et al. 1988; Putnam 2000; Singh-Manoux and Marmot 2005). Similarly, the economics literature has often pointed out the positive externalities of synchronicity both in leisure and in market work (e.g., Weiss 1996). Few studies have tried to identify exogenous determinants of coordination. For example, public holidays have been found to be welfare enhancing, not only by increasing the amount of leisure to each individual, but also by increasing the coordination of leisure activities among individuals (e.g., Mers and Osberg 2006). Similarly, Hamermesh et al. (2008) found that an exogenous shock to time in one area because of daylight saving time lead its residents to change their work schedule to coordinate their other (leisure) activities with those in adjacent areas.
The third class of indicator is leisure activity fragmentation. For a given amount of leisure time, individuals with more fragmented leisure may be justifiably more rushed and stressed. To measure the fragmentation of leisure, we use the number of leisure intervals during the diary day. An interval is defined as an uninterrupted period of time when the individual is engaged in one of these four main activities: market work, personal care, home production, and leisure. We then define a leisure interval as that time interval in which the main activity is leisure (regardless of whether the interval contains two or more different leisure episodes).10
The number of leisure intervals can thus give only a partial picture of leisure fragmentation. In particular, the difference in the change in the number of leisure intervals for low-educated and highly educated individuals does not necessarily imply a difference in the fragmentation of leisure between the two groups because leisure time evolved differently for each group. To address this caveat, we also report the average duration of leisure intervals, defined as the individual’s amount of leisure (in minutes per day) divided by the individual’s number of leisure intervals.
It is important to note that the switch of method from the diary that the diarist sees (1965, 1975, and 1985) to the telephone diary (1993 and 2003) seems to have resulted in a decline in the number of episodes in recent surveys. Also, for the earlier studies (1965, 1975, and 1985), respondents were automatically assigned a new episode with a change of location, main activity, or secondary activity, which results in these surveys having more episodes. To the extent that low-educated and highly educated individuals are affected by data collection methods in the same way, this artificial decrease in the mean number of episodes over the five surveys should not affect the relative trends in leisure quality reported here. Nonetheless, we acknowledge that absolute trends in the number of leisure intervals and the duration of leisure intervals over this period might also be capturing survey design changes as well as genuine changes in the fragmentation of activities. We thus define the normalized number of intervals as the number of intervals divided by the total number of intervals in the survey, and the normalized average duration of leisure intervals as the total amount of leisure divided by the normalized number of intervals. We discuss additional evidence from these normalized measures in the next section.
Table 1 shows descriptive statistics for the indicators of quantity and quality of leisure over the sample period for all men and women, and by education group, separately.11 We perform the analysis for highly educated and low-educated individuals separately. A highly educated individual is defined as having more than a high school diploma or GED equivalent: that is, as having some college or a college degree or more (13 or more years of schooling). Low-educated individuals are those with up to 12 years of schooling: that is, having at most a high school diploma.
The first row in Panels A and B of Table 1 shows trends in the quantity of leisure that are similar to those found in previous research.12 Men’s average hours of leisure per week exhibit a statistically significant increase over the period of reference, from 28 hours of leisure per week in 1965 to 33 hours and 30 minutes of leisure per week in 2003. Women’s leisure time follows a similar pattern, increasing (on average) by 3 hours and 30 minutes per week over the relevant period from 27 hours in 1965. The increases in leisure time are greater for low-educated individuals. At the beginning of the period, low-educated men started with 3 hours and 20 minutes more of leisure than highly educated men, and low-educated women started with 50 minutes more of leisure than highly educated women. By the end of the period, however, the differences between the two education groups had widened: low-educated men and women had, respectively, 3 hours and 35 minutes, and 3 hours and 45 minutes more leisure than highly educated men and women.
Between 1965 and 1985, the percentage of pure leisure decreased by an average of 5 and 5.7 percentage points, respectively, for men and women. There was also a statistically significant decrease in the percentage of leisure with spouse for men between 1965 and 2003. Married men reported spending less time in the presence of the spouse over this period, reducing from an average of 57 percentage points in 1965 to 53 percentage points in 2003. In contrast, married women reported spending more leisure time in the presence of the spouse over this period, increasing from an average of 44.5 percentage points in 1965 to 49.5 percentage points in 2003.13 The percentage of leisure with adults decreased for men and women by an average of 13 and 7 percentage points, respectively, between 1965 and 2003. Although raw figures show a general decrease in the number of leisure intervals for women, as well as an increase in the average duration of leisure intervals for both men and women, they seem to be masking the real increase in the fragmentation of leisure. After we adjust for changes in the total number of intervals, which we believe likely is an artifact of changes in survey methodology, there is indeed an increasing trend in the normalized number of intervals for both men and women, and a decline in the normalized average duration of leisure intervals for women.14
Thus, in stark contrast with the changing amount of leisure, most of our quality indicators show declines in the quality of leisure time over this period for both men and women. Moreover, although low-educated men and women started off with higher quality of leisure according to most of our indicators (with the exception of pure leisure), the differences between the two education groups narrowed over this period because the quality of leisure decreased differentially more for low-educated than for highly educated individuals. These trends, however, do not take into account sample composition effects. The average American in 2003 is more likely to be highly educated, to be single, and to have fewer children than those in 1965. All these changes may affect how an individual chooses to allocate his or her time, and thus controlling for demographics is also important for the analysis of the trends of the quality and quantity of leisure over time. The next section explores these trends across educational groups to examine the extent to which the quantity and quality of leisure have become more unequal between education groups, adjusted for demographic changes.
Trends in the Quantity and Quality of Leisure by Educational Status
The superscript e represents our education categories. We perform the analysis for highly educated and low-educated individuals separately, as defined in the previous section. The coefficient of interest is the T year dummy coefficient () for each educational category e, where T represents the last year for which we have information on the dependent variable; this year is 2003 for all leisure quality indicators except for our indicator of pure leisure, for which T is 1985. These coefficients inform us about how the quantity and quality of leisure have changed over time for each educational group, controlling for changes in key demographics.
A positive value of when the dependent variable is the amount of leisure would indicate increases in leisure over these decades. A negative coefficient when the dependent variables are our indicators of pure leisure, co-present leisure, and average duration of leisure intervals, and positive when our dependent variable is the number of leisure intervals, would indicate decreasing quality of leisure in year T with respect to 1965. To answer the question of whether low-educated individuals have lost leisure quality relative to more-educated individuals over this period, we compare changes in our leisure quality metrics for low-educated workers and highly educated workers (i.e., ), and also check whether the difference (Diff) is statistically significant.
Panel A and Panel B in Table 2 show the trends in the quantity and quality of leisure for men and women, controlling for demographic characteristics. Column 1 shows the coefficient for low-educated individuals, and column 2 shows the coefficient for individuals with at least some college education. Column 3 reports the difference between these two coefficients, , and informs us about the direction of the changes in leisure time and leisure quality for each education category during the relevant period. (The relevant period is 1965–2003 for all indicators, with the exception of the percentage of pure leisure, which is observed only between 1965 and 1985). The p value of this difference is reported in column 4 and indicates whether there is a statistically significant difference in the trends in the quantity and quality of leisure between the two educational groups. (The remaining year coefficients included in the regression are available on request.)
The general picture that emerges from Table 2 is that whereas individuals with less than 13 years of schooling increased the quantity of leisure with respect to 1965 more than the college-educated group, the relative quality of leisure for low-educated individuals deteriorated over time. The difference in the trends between the two educational groups is generally statistically significant (column 4).
The first row of Panel A in Table 2 shows that for men, the 1965–2003 increase in leisure time for low-educated individuals is statistically significant and accounts for almost 5 hours and 15 minutes per week, as indicated by the coefficient on leisure time in the year 2003 in column 1. Low-educated men increased leisure time by almost 1 hour more than highly educated men (column 3), and this difference is statistically significant at the 95% level, as shown in column 4. Panel B shows a similar picture for women, although the differences across educational groups are greater. In 2003, low-educated women enjoyed 5.5 more hours per week of leisure time than in 1965, whereas highly educated women had an increase of 2 hours of leisure per week. The difference of 3.4 hours (column 3) between educational groups is statistically significant at the 99% level, as shown in column 4.
The remaining rows in columns 1 and 2 in Panel A of Table 2 show that highly educated men experienced smaller declines in the quality of leisure than less-educated men. Men with some college or more experienced a statistically significant decrease of 3.5 percentage points in the percentage of pure leisure between 1965 and 1985. Although they did not experience a statistically significant change in the percentage of leisure with spouse, they had a statistically significant decrease in the percentage of leisure with adults of 7 percentage points between 1965 and 2003. In contrast, low-educated men experienced much larger decreases in all indicators: namely a decrease of 6 percentage points in pure leisure and leisure with spouse, and a decrease of 10.5 percentage points in leisure with adults. Given that the average pure leisure time over the relevant period is of 25 hours and 45 minutes per week for low-educated men, these percentages translate into losses in pure leisure of 1 hour and 35 minutes (25.71 × 6.33 = 1.62) per week. Similarly, the coefficients on leisure with spouse and leisure with adults suggest decreases of the order of 1 hour (17.35 × 5.97) and 2 hours and 15 minutes (21.39 × 10.54) per week, respectively. (The average leisure time with the spouse and leisure with other adults over the relevant period for low-educated men is 17 hours and 20 minutes, and 21 hours and 25 minutes per week, respectively). For highly educated men, these percentages translate into a loss of 50 minutes of pure leisure (23.54 × 3.54) and 1 hour and 20 minutes (19.07 × 6.81) of leisure with adults. (The average pure leisure and leisure with other adults over the relevant period for highly educated men is 23 hours and 30 minutes, and 19 hours per week, respectively.)
Results from raw and normalized data suggest that highly educated men have become relatively better off in terms of the fragmentation of leisure than low-educated men. Highly educated men experienced a decrease in the number of leisure intervals of 0.11 leisure intervals between 1965 and 2003, but low-educated men did not experience any statistically significant change. The average duration of leisure intervals also increased more for highly educated individuals (20 minutes per day) versus an increase of just 18 minutes per day for low-educated men (although this difference is statistically significant only at the 90% level). Similarly, relative trends using a normalized measure of leisure intervals show that increases in the normalized number of intervals were lower for highly educated men than for low-educated men and that the normalized duration of leisure intervals increased for highly educated individuals but decreased for low-educated individuals.
The relative quality of leisure also declined more for low-educated women than for women with some college or more. Low-educated women experienced a statistically significant decrease in the percentage of pure leisure of almost 7 percentage points between 1965 and 1985, whereas highly educated women experienced a decrease of just 4.4 percentage points. Because the average pure leisure time is 27 hours and 5 minutes per week and 24 hours and 30 minutes per week, respectively, for low-educated and highly educated women over this period, these coefficients translate into a reduction of 1 hour and 50 minutes (27.09 × 6.80) of pure leisure per week for low-educated women, and of 1 hour (24.47 × 4.35) of pure leisure per week for highly educated women. Restricting the sample to married women shows increases in the percentage of leisure with spouse for both educational groups between 1965 and 2003. However, highly educated women experienced a larger increase in this indicator than low-educated married women (8 vs. 3 percentage points). These coefficients suggest increases in leisure with spouse of 25 minutes (15.14 × 2.74) per week for low-educated women and 1 hour and 10 minutes (13.94 × 8.11) per week for highly educated women. The percentage of leisure with adults also decreased slightly more for the highly educated group.
Although highly educated women experienced a greater decrease in the number of leisure intervals, the smaller increase in leisure with respect to low-educated women resulted in lower increases in the average duration of leisure intervals for highly educated women. Thus, highly educated women experienced a higher fragmentation in leisure than low-educated women. Although highly educated women experienced a decrease in the number of leisure intervals of 0.35 intervals, low-educated women experienced a decrease of just 0.26 intervals—a difference that is statistically significant at the 99% level. However, the average duration of leisure intervals increased by 26 minutes per day for low-educated women over this period versus an increase of just 15 minutes per day for highly educated women; this difference is statistically significant at the 99% level. Using a normalized measure of the average duration of leisure yields the same conclusion.
Female labor force participation increased substantially over the observation period. To the extent that increases in working hours are correlated with educational status, omitting employment status in our analysis may lead to a bias in our estimates. Panel C in Table 2 shows results for working women. The 1965–2003 increase in leisure time for working women with some high school and high school graduates is statistically significant and accounts for almost 7 hours per week; the increase for the college-educated group is much smaller, at 5 hours per week. Column 4 shows that the two-hour difference in the leisure quantity trends between educational groups is statistically significant at the 99% level. Results on the quality of leisure for working women across educational groups are, however, mixed. Highly educated working women did not experience a statistically significant decrease in the percentage of pure leisure, although low-educated working women experienced a decline of almost 8 percentage points, resulting in a decrease in the amount of pure leisure of 1 hour and 45 minutes (22.51 × 7.70). (Between 1965 and 1985, average pure leisure was 22 hours and 30 minutes for low-educated working women.)
Similarly to the results for all women, the probability of highly educated working women spending leisure time with the spouse increased to a much greater extent than that for low-educated working women between 1965 and 2003 (10.5 percentage points and 4 percentage points, respectively). However, highly educated working women are less likely than low-educated working women to be spending leisure time with other adults (3.5 percentage points and 1 percentage point, respectively). Working women are the only group for whom the number of leisure intervals increased over the period. However, highly educated working women experienced a smaller increase in the number of leisure intervals (0.32 versus 0.15 additional leisure intervals for low-educated and highly educated working women, respectively). Similarly to all women (i.e., working and nonworking), however, the increase in the average duration of leisure intervals was much higher for low-educated working women (an increase of 21 minutes per day for low-educated working women versus an increase of 14 minutes per day for highly educated women; significant at the 99% level).
The Role of Television
Time spent watching television has been one of the most important contributors to the increase in leisure during this period for both low-educated and highly educated individuals, and represents the highest proportion of total leisure time (44% of leisure time for highly educated individuals and 52% of leisure time for low-educated individuals). To the extent that watching television can be considered low-quality leisure, the fact that the time devoted to this activity has increased slightly more for low-educated individuals over this period may explain why the quality of leisure declined more for low-educated than for highly educated adults.16
To test this alternative explanation, we undertake the analysis developed in the previous section, omitting the time spent watching television. Table 3 shows that removing television watching from the analysis results in a relatively flat pattern of leisure over this period for men but a significant decrease of leisure for women. However, the apparent decline in leisure without television time is significantly higher for highly educated individuals. Thus, although the increase in the time spent watching television seems responsible for the increases in leisure over this period, low-educated individuals continue to experience a more favorable trend in leisure terms than highly educated individuals even after the time spent watching television is removed; thus increases in the time watching television cannot explain the differential trend in the quantity of leisure between educational groups.
Increases in the time spent watching television cannot entirely explain the differential trend in leisure in terms of quality, either. The decline in leisure quality for all educational groups over this period continues to hold for the most part even after the time spent watching television is omitted. Interestingly, the time spent with the spouse increases for highly educated men when the time spent watching television is excluded from the analysis, whereas low-educated men and women decrease leisure with spouse to a greater extent after television is excluded from the analysis. However, the time spent watching television can account for the differential trend between the two educational groups in leisure with adults for men but not for the variation in leisure quality between educational groups for women. Other leisure activities, such as at-home leisure and read/listen, which also represent a significant fraction of the total amount of leisure time, similarly fail to explain the differential trend in leisure quality across educational groups.
This article moves beyond previous research, which has mostly focused on the total amount of time devoted to leisure, and begins to provide a more comprehensive view of how leisure inequality across educational groups has evolved over the period 1965–2003 in the United States. We use the richness of the diary information in the American Heritage Time Use Study (AHTUS) to construct several indicators of the quality of leisure time: pure leisure, co-present leisure, and leisure fragmentation. Consistent with previous studies, we find general increases in leisure time across educational groups between 1965 and 2003, especially for low-educated individuals. However, although the quality of leisure decreased over the period, this decline was greatest for low-educated adults.
A possible explanation of the unequal distribution of leisure quantity and quality may be a greater decrease in the relative price of quality leisure for highly educated individuals than for low-educated individuals over this period. There is evidence that highly educated individuals use their earning power to work at more desirable times (despite working longer hours) than low-educated individuals (see Hamermesh 1999, 2002), which may enable highly educated individuals to time their leisure in order to make it less fragmented, and to coordinate it with others’ leisure, even if in exchange for a lower quantity of leisure. We leave a more thorough investigation of these important questions for further research.
Conclusions in this research are those drawn by the authors and may not reflect the views of the creators or funders of AHTUS or the collectors of the original surveys harmonized in this data set. We are grateful for the financial support provided by the Economic and Social Research Council (Grant Number RES-060-25-0037) and the Spanish Ministry of Education and Science (Project ECO2008-01297), and for helpful comments from three anonymous referees.
Appendix 1: Technical Information
For the sake of comparison with previous studies, and to minimize the role of time-allocation decisions (such as education and retirement) that have a strong intertemporal component over the life cycle, we restrict the sample used throughout the analysis to nonretired/nonstudent individuals aged 21–65; results should be interpreted as being “per working-age adult” (or per adult within the specified subsample, when relevant). This approach also avoids possible biases from the changing proportion of retired individuals in the general population over this long period.
Not including individuals out of the labor force may be particularly problematic if low-educated individuals are more likely to be non-employed and thus have more leisure time. This is a problem that we share with previous studies that looked at trends in the amount of leisure. The fact that we control for total leisure time in our analysis makes this issue less problematic for our indicators of leisure quality, although possible biases remain if low-educated individuals are more likely to be retired and if leisure time during retirement differs in quality from that during the active working years. We have conducted robustness checks including retired and nonretired individuals aged 24–65 and aged 24–72. Results do not change (available on request).
We also restrict the sample to include only individuals who have time diaries that add up to a complete day (1,440 minutes) and whose diaries are not “low quality” (i.e., they have 90 or fewer minutes missing for main activity time), have seven or more episodes, and have some time recorded in at least three of four basic activities (sleep or rest, eat or drink, personal care, and travel) either as a primary or secondary activity.17 The excluded diaries represent 10% of the sample aged 21–65, and results are robust to their inclusion.
We further limit the sample to married individuals or those individuals living with a partner wherever the dependent variable is the quality indicator percentage of leisure time with the spouse. This sample restriction is necessary because trends in marriage rates and the timing of marriage have changed over time, especially for highly educated individuals; and if marriage patterns alter behavior in daily routines, such as time together, they could in principle explain some patterns in the data.
Appendix 2: Validating the Four Classes of Indicators
Here we validate our leisure quality indicators using the 1985 element of the AHTUS, which collected an additional item of information not available elsewhere in the sequence of surveys: an activity enjoyment “rating” (on a 0–10 Dislike It/Like It scale) attached to each event (see Robinson 1993). The 1985 AHTUS did not include, however, the “with whom?” diary information for each registered event. We use a similar diary data set (although rating activities on a 1–5 Like It/Dislike It scale, and collecting information on a fixed 30-minute grid, rather than the open intervals used in the U.S. survey), from a national random sample of individuals living as members of heterosexual couples in the United Kingdom in 1986, which does include co-presence data (see Sullivan 1996a,b).
The vector Ii,j contains four leisure quality indicators as described in the second section of the article. The vector Xi includes socioeconomic variables of the individual, which include age, age squared, gender, an indicator variable for part-time and full-time work, an indicator variable that takes the value of 1 if there is a child younger than 5 at home, and another indicator variable that takes the value of 1 if there is a child between 5 and 18 years old living in the household. The vector Ai,j includes six dummy variables indicating the nature of the leisure activity being done. These are classified into out of home leisure, active sport and exercise, read and listening to music, watch television (reference activity), other leisure at home, and writing.
Panels A and B in Table 5 show the associations between the leisure quality indicators and the enjoyment scores in the U.S. and UK data, respectively. In all specifications, it emerges that these indicators are all associated with the activity enjoyment ratings in the expected way. The fragmentation effects, meanwhile, are hardly affected by the additional variables.
Costa (2000) found a similar pattern, documenting that low-wage workers reduced their market work hours relative to high-wage workers between the 1890s and 1991.
One-quarter of time that would be considered “leisure,” according to the conventional implementation of the third-person criterion, and one-third of what would conventionally be considered “work,” is unexpectedly placed by the diarists (Gershuny 2009b).
Results for the other definitions of leisure are available upon request.
Although some activities coded in AHTUS were not coded as such in the original surveys (because the coding of activities was different), we have paid particular attention to having comparable activities defined for the different years. The only exception is computer use, which is not coded as such in surveys for 1965 and 1975, reflecting the fact that few Americans could afford, or had the technology acumen to use, a personal computer. Excluding computing from our definition of leisure would not have been an option because leisure in the latter years would have been underreported. For more information, refer to the Concordance Files at http://www.timeuse.org/files/cckpub/AHTUS-USAConcordanceFiles-20091202.xls and the Variables in the Diary Files table at http://www.timeuse.org/files/cckpub/AHTUS-Variable-In-Time-Diary-Files-20100109.pdf.
Experience sampling was developed to collect information on people’s reported feelings in real time in natural settings during selected moments of the day (Csikszentmihalyi 1990; Stone and Schiffman 1994). Under this method, participants carry a handheld computer that prompts them several times during the course of the day (or days) to answer a set of questions immediately, such as questions asking about their physical location, the activities in which they were engaged just before they were prompted, or the people with whom they were interacting. They also report their current subjective experience by indicating the extent to which they feel the presence or absence of various feelings, such as feeling angry, happy, tired, or impatient (Kahneman and Krueger 2006; Steptoe et al. 2005).
The day reconstruction method has been used, for example, in the collection of the Princeton Affect and Time Survey (PATS). Here, respondents were asked to reconstruct the previous day by completing a short diary. Then, three 15-minute intervals were randomly selected from the nonsleeping portion of the diary, and respondents were asked the extent to which they experienced six different feelings (pain, happy, tired, stressed, sad, and interested) during each interval (Krueger 2007). Previously, the otherwise similar yesterday-diary approach was used to collect information on the levels of instant enjoyment for all the episodes (not just three) in the diary as in the examples from the United Kingdom and the United States AHTUS in Appendix 2. Unlike the PATS, these surveys collected one dimension of instant enjoyment, which is scaled from 5 to 0 and 0 to 10, respectively. Knabe et al. (2010) used both types of measures and reached the same conclusions with the two types of measures.
An alternative method would require imputations of enjoyment levels for the other survey years (either at the activity level, as in Krueger (2007), or at the individual level). A potential limitation to this method (see Krueger 2007) is that it maintains the nature of activities relatively constant, not only over time, but also across educational groups. This latter point is particularly relevant in the current context because different groups of individuals may rank the same activity differently, and the mix of these responses may change over time. Results from our validation exercise in Appendix 2 suggest that our indicators can still be used as a good proxy for leisure quality. We thus leave investigating this alternative method for future research.
Extensive evidence points toward positive assortative mating along education (Blossfeld and Timm 2003; Lam 1988), and thus perhaps highly educated individuals have a higher preference for spending leisure time with a spouse, precisely because the spouse is also highly educated. This hypothesis does not seem to be ratified by the results shown in Appendix 2, however. We find that individual’s enjoyment of leisure time when accompanied by the spouse is greater, regardless of educational class.
The category “other adult” is considered to be the spouse or partner, another adult from the household, a shop or professional worker, a coworker, a person well known, and other (adult) person present. Unfortunately, the AHTUS lacks comparable information across years on whether a child is present, and thus comparisons along these lines are not possible.
The diary survey is organized in episodes. Thus, two consecutive leisure activities are considered to be different episodes (i.e., reading and cycling), but these consecutive leisure episodes are considered to be the same leisure interval according to our definition.
We use the weights provided in the AHTUS. These weights account for population/sample distribution by age group and sex, and provide an even distribution of the days of the week. All cases with missing basic information or bad diaries are 0-weighted and thus are excluded from the analysis. Further information on these weights can be found in the AHTUS codebook (http://www.timeuse.org/ahtus/documentation/docs/pdf/Codebook.pdf).
Aguiar and Hurst (2007) found that leisure increased by 6–8 hours per week for men and 4–8 hours per week for women over the period 1965–2003. Burda et al. (2008) found that the amount of leisure time between 1985 and 2003 decreased by 13.3 and 2.7 minutes per day for men and women, respectively. Differences between the two sets of results might be due to the sample used in their analysis, which was a sample of individuals aged 20–74.
The discrepancy in the reported leisure with spouse between married men and women is due to the fact that married women are more likely to do housework while their spouses are enjoying leisure. Fisher et al. (2007) illustrated the persistence of gender differences in the time when partners are together: when there is unpaid labor to be done during leisure time, women still remain more likely to carry out the chores (e.g., while he sits in front of the television, perhaps talking with her while she sets the table and finishes food preparation). Similarly, the different trends in this indicator between men and women are the result of women’s decrease in the time devoted to household chores over these decades.
The number of normalized intervals increased by 1.21 for men and by 3.62 for women, whereas the normalized duration of intervals increased for men by 1.35 minutes per day for men but decreased by 0.82 minutes per day for women.
In all years except 1993, the time-use surveys asked respondents to report their marital status. Although our base results do not include this control (because they are unavailable for 1993), we reran all of our regressions, including marital status as an additional control, on a sample that excludes the 1993 survey. This modification did not alter the main findings of our article, and results are available upon request.
Television watching can be considered to be low-quality leisure. The literature reports that television watching is a more passive or one-way communication medium (see Kubey and Csikszentmihalyi 1990), and evidence based on diary-affect data and general satisfaction surveys suggests that television watching is not as enjoyable as other leisure activities. Consistent with what is shown in Appendix 2, television watching ranks low in the level of instant enjoyment with respect to other leisure activities (Kahneman et al. 2004; Robinson and Godbey 1997). Substantial evidence also shows that television watching is negatively linked with life satisfaction in general (Espe and Seiwert 1987; Frey and Stutzer 2007; Kasser 2002; Morgan 1984; Putnam 1995, 2000; Shrum et al. 2003; Tankard and Harris 1990). Authors’ calculations show that television watching is indeed less often enjoyed with other adults (61% of the time as opposed to 65% for the rest of leisure activities) and it is also a more fragmented type of leisure.
Some respondents providing childcare for multiple children or an infant, as well as some diarists performing adult care, did not record travel, and also missed a second or third basic activity. If these diaries from caregivers nonetheless included at least 10 episodes, we counted these diaries as good diaries; it may be possible that the diarists ate while feeding the care recipient, for example, but did not record their own eating.
The question that these models are attempting to address is how well the leisure quality proxies explain the variation in enjoyment across leisure activities. Thus, we restrict our analysis to leisure activities because including all the activities (work, leisure, and sleep) would capture the proportion of explained variation in enjoyment across all activities—not the proportion across leisure activities—and some significant share of that explained variation would be accounted for by activity type variables rather than the proxies for leisure quality.